Indicator Heterogeneity I: Temporal Heterogeneity
Addison
PARCOMPUTE = TRUE
N_CORE = parallel::detectCores()Background
In this notebook, we repeat the analysis of 02_temporal_heterogeneity.Rmd for all of our core indicators.
Data setup
# Fetch the following sources and signals from the API
# TODO: Add Google Symptoms "eventually"
source_names = c("doctor-visits", "fb-survey", "fb-survey", "hospital-admissions")
signal_names = c("smoothed_adj_cli", "smoothed_cli", "smoothed_hh_cmnty_cli",
"smoothed_adj_covid19")
pretty_names = c("Doctor visits", "Facebook CLI", "Facebook CLI-in-community",
"Hospitalizations")
target_names = c("Cases", "Cases", "Cases", "Deaths")
geo_level = "county"
start_day = "2020-04-15"
end_day = NULL
cache_fname = 'cached_data/03_heterogeneity_core_indicators.RDS'
if (!file.exists(cache_fname)) {
df_signals = vector("list", length(signal_names))
for (i in 1:length(signal_names)) {
df_signals[[i]] = suppressWarnings(
covidcast_signal(source_names[i], signal_names[i],
start_day, end_day,
geo_type=geo_level))
}
# Fetch USAFacts confirmed case incidence proportion (smoothed with 7-day
# trailing average)
df_cases = suppressWarnings(
covidcast_signal("usa-facts", "confirmed_7dav_incidence_prop",
start_day, end_day,
geo_type=geo_level))
df_deaths = suppressWarnings(
covidcast_signal("usa-facts", "deaths_7dav_incidence_prop",
start_day, end_day,
geo_type=geo_level))
saveRDS(list(df_signals, df_cases, df_deaths), cache_fname)
} else {
cached_data = readRDS(cache_fname)
df_signals = cached_data[[1]]
df_cases = cached_data[[2]]
df_deaths = cached_data[[3]]
}
case_num = 500
geo_values = suppressWarnings(covidcast_signal("usa-facts", "confirmed_cumulative_num",
max(df_cases$time_value),
max(df_cases$time_value))) %>%
filter(value >= case_num) %>% pull(geo_value)## Fetched day 2020-11-09: 1, success, num_entries = 3192geo_values = suppressWarnings(covidcast_signal("usa-facts", "confirmed_cumulative_num",
'2020-11-01',
'2020-11-01')) %>%
filter(value >= case_num) %>% pull(geo_value)## Fetched day 2020-11-01: 1, success, num_entries = 3192Setup
sensorize_time_ranges = list(
c(-7, -1),
c(-10, -1),
c(-14, -1),
c(-21, -1))
# TODO: Add more "core indicators"
for (ind_idx in 1:length(source_names)) {
if (target_names[ind_idx] == 'Cases') {
df_target = df_cases
} else if (target_names[ind_idx] == 'Deaths') {
df_target = df_deaths
} else {
stop(sprintf("No matching dataframe for target %s.", target_names[ind_idx]))
}
ind_df = tibble(df_signals[[ind_idx]]) %>% filter(geo_value %in% geo_values)
ind_target = inner_join(ind_df, tibble(df_target),
by=c('geo_value', 'time_value')) %>% select (
geo_value=geo_value,
time_value=time_value,
indicator_value=value.x,
target_value=value.y,
)
ind_global_sensorized = ind_target %>% group_by (
geo_value,
) %>% group_modify ( ~ {
fit = lm(target_value ~ indicator_value, data =.x);
tibble(time_value=.x$time_value,
indicator_value=.x$indicator_value,
target_value=.x$target_value,
sensorized_value=fit$fitted.values)
}) %>% ungroup
df_global_sensorized = ind_global_sensorized %>% transmute (
geo_value=geo_value,
signal='ind_sensorized',
time_value=time_value,
direction=NA,
issue=lubridate::ymd('2020-11-01'),
lag=NA,
value=sensorized_value,
stderr=NA,
sample_size=NA,
data_source='linear_sensorization',
)
attributes(df_global_sensorized)$geo_type = 'county'
attributes(df_global_sensorized)$metadata$geo_type = 'county'
class(df_global_sensorized) = c("covidcast_signal", "data.frame")
base_cor_fname = sprintf('results/03_base_cors_%s_%s.RDS',
source_names[ind_idx], signal_names[ind_idx])
if (!file.exists(base_cor_fname)) {
df_cor_base_ind = covidcast_cor(df_signals[[ind_idx]], df_target,
by='time_value', method='spearman')
df_cor_sensorized_ind = covidcast_cor(df_global_sensorized, df_target,
by='time_value', method='spearman')
df_cor_base = rbind(df_cor_base_ind, df_cor_sensorized_ind)
df_cor_base$Indicator = as.factor(c(rep('Raw', nrow(df_cor_base_ind)),
rep('Sensorized (Spatial)',
nrow(df_cor_sensorized_ind))))
saveRDS(df_cor_base, base_cor_fname)
} else {
df_cor_base = readRDS(base_cor_fname)
}
sensorize_fname = sprintf('results/03_sensorize_cors_%s_%s.RDS',
source_names[ind_idx], signal_names[ind_idx])
sensorize_val_fname = sprintf('results/03_sensorize_vals_%s_%s.RDS',
source_names[ind_idx], signal_names[ind_idx])
if (!file.exists(sensorize_fname)) {
sensorize_cors = vector('list', length(sensorize_time_ranges))
ind_target_sensorized_list = vector('list', length(sensorize_time_ranges))
for (outer_idx in 1:length(sensorize_time_ranges)) {
sensorize_llim = sensorize_time_ranges[[outer_idx]][1]
sensorize_ulim = sensorize_time_ranges[[outer_idx]][2]
min_sensorize_date = lubridate::ymd(start_day) - sensorize_llim
max_sensorize_date = max(ind_target$time_value)
sensorize_date_offsets = 0:(max_sensorize_date-min_sensorize_date)
joiner_df_list = vector('list', length(sensorize_date_offsets))
for (idx in 1:length(sensorize_date_offsets)) {
dt = sensorize_date_offsets[idx]
sensorize_date = min_sensorize_date + dt
joiner_df_list[[idx]] = tibble(
sensorize_date = sensorize_date,
time_value = sensorize_date + sensorize_llim:sensorize_ulim)
}
joiner_df = bind_rows(joiner_df_list)
if (!PARCOMPUTE) {
ind_sensorized_lm = ind_target %>% full_join(
joiner_df,
on='time_value',
) %>% group_by (
geo_value,
sensorize_date,
) %>% group_modify (
~ broom::tidy(lm(target_value ~ indicator_value, data = .x))
) %>% ungroup
} else {
ind_grouped_list = ind_target %>% full_join(
joiner_df,
on='time_value',
) %>% group_by (
geo_value,
sensorize_date,
) %>% group_split
ind_sensorized_lm = parallel::mclapply(ind_grouped_list, function(df) {
broom::tidy(
lm(target_value ~ indicator_value, data = df)
) %>% mutate (
geo_value = unique(df$geo_value),
sensorize_date = unique(df$sensorize_date),
)}, mc.cores = N_CORE) %>% bind_rows
}
ind_sensorized_wide = ind_sensorized_lm %>% select(
geo_value,
sensorize_date,
term,
estimate,
) %>% mutate (
term = sapply(term, function(x) {ifelse(x=='(Intercept)',
'intercept',
'slope')}),
) %>% pivot_wider (
id_cols = c(geo_value, sensorize_date),
names_from=term,
values_from=estimate,
)
ind_target_sensorized = ind_target %>% inner_join (
ind_sensorized_wide,
by=c('time_value'='sensorize_date',
'geo_value'),
) %>% mutate (
sensorized_value = intercept + indicator_value * slope,
)
df_sensorized = ind_target_sensorized %>% transmute (
geo_value=geo_value,
signal='ind_sensorized',
time_value=time_value,
direction=NA,
issue=lubridate::ymd('2020-11-01'),
lag=NA,
value=sensorized_value,
stderr=NA,
sample_size=NA,
data_source='linear_sensorization',
)
attributes(df_sensorized)$geo_type = 'county'
class(df_sensorized) = c("covidcast_signal", "data.frame")
df_cor_sensorized_ind = covidcast_cor(df_sensorized, df_target,
by='time_value', method='spearman')
df_cor_sensorized_ind$Indicator = sprintf('Sensorized (TS, %d:%d)',
sensorize_llim,
sensorize_ulim)
sensorize_cors[[outer_idx]] = df_cor_sensorized_ind
ind_target_sensorized_list[[outer_idx]] = ind_target_sensorized
}
saveRDS(sensorize_cors, sensorize_fname)
saveRDS(ind_target_sensorized_list, sensorize_val_fname)
} else {
sensorize_cors = readRDS(sensorize_fname)
}
df_cor = bind_rows(df_cor_base, sensorize_cors)
df_cor$Indicator = factor(df_cor$Indicator,
levels=c('Raw',
'Sensorized (Spatial)',
sapply(sensorize_time_ranges,
function(x) {
sprintf('Sensorized (TS, %d:%d)',
x[[1]], x[[2]])
})))
plt = ggplot(df_cor, aes(x = time_value, y = value)) +
geom_line(aes(color = Indicator)) +
labs(title = sprintf("Correlation between %s and %s",
pretty_names[ind_idx],
target_names[ind_idx]),
subtitle = "Per day",
x = "Date", y = "Correlation") +
theme(legend.position = "bottom")
print(plt)
}## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"## Warning: Removed 64 row(s) containing missing values (geom_path).## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Warning: Removed 52 row(s) containing missing values (geom_path).## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Warning: Removed 52 row(s) containing missing values (geom_path).## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Joining, by = "time_value"
## Warning: Removed 310 row(s) containing missing values (geom_path).
QUANTS = c(0.01, 0.99)
# TODO: Add more "core indicators"
for (ind_idx in 1:length(source_names)) {
if (target_names[ind_idx] == 'Cases') {
df_target = df_cases
} else if (target_names[ind_idx] == 'Deaths') {
df_target = df_deaths
} else {
stop(sprintf("No matching dataframe for target %s.", target_names[ind_idx]))
}
base_cor_fname = sprintf('results/03_base_cors_%s_%s.RDS',
source_names[ind_idx], signal_names[ind_idx])
sensorize_fname = sprintf('results/03_sensorize_cors_%s_%s.RDS',
source_names[ind_idx], signal_names[ind_idx])
sensorize_val_fname = sprintf('results/03_sensorize_vals_%s_%s.RDS',
source_names[ind_idx], signal_names[ind_idx])
df_cor_base = readRDS(base_cor_fname)
sensorize_cors = readRDS(sensorize_fname)
sensorized_vals = readRDS(sensorize_val_fname)
for (inner_idx in 1:length(sensorize_time_ranges)) {
sv = sensorized_vals[[inner_idx]]
print(summary(sv$slope))
print(slope_limits <- quantile(sv$slope, QUANTS, na.rm=TRUE))
plt = ggplot(
sensorized_vals[[inner_idx]],
aes(x=time_value,
y=slope),
) + geom_point (
alpha=0.1,
size=0.5,
) + geom_hline (
yintercept=0,
colour='white',
) + stat_summary (
aes(y=slope,
group=1,
colour='median'),
fun=median,
geom="line",
group=1,
) + stat_summary (
aes(y=slope,
group=1,
colour='+/- mad'),
fun=function(x) { median(x) + mad(x) },
geom="line",
group=1,
) + stat_summary (
aes(y=slope,
group=1,
colour='+/- mad'),
fun=function(x) { median(x) - mad(x) },
geom="line",
group=1,
) + scale_colour_manual(
values=c("median"="maroon",
"+/- mad"="darkgreen")
) + labs(
colour=''
) + ggtitle(
sprintf("Slope distribution for %s, fitted on t in %d:%d",
pretty_names[ind_idx],
sensorize_time_ranges[[inner_idx]][1],
sensorize_time_ranges[[inner_idx]][2])
) + ylim (
slope_limits[[1]], slope_limits[[2]]
)
print(plt)
}
}## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -3488425 -1 0 4078 2 574876473 4382
## 1% 99%
## -25.35813 27.09141## Warning: Removed 10364 rows containing non-finite values (stat_summary).
## Warning: Removed 10364 rows containing non-finite values (stat_summary).
## Warning: Removed 10364 rows containing non-finite values (stat_summary).## Warning: Removed 10364 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -3693626 -1 0 4675 2 582794882 3340
## 1% 99%
## -19.93872 22.74537## Warning: Removed 9260 rows containing non-finite values (stat_summary).## Warning: Removed 9260 rows containing non-finite values (stat_summary).
## Warning: Removed 9260 rows containing non-finite values (stat_summary).## Warning: Removed 9260 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -3693626 -1 0 1125 2 52206253 2494
## 1% 99%
## -15.93316 19.67603## Warning: Removed 8320 rows containing non-finite values (stat_summary).## Warning: Removed 8320 rows containing non-finite values (stat_summary).
## Warning: Removed 8320 rows containing non-finite values (stat_summary).## Warning: Removed 8320 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -3693626 -1 0 1432 2 53331161 1675
## 1% 99%
## -12.33914 17.12680## Warning: Removed 7313 rows containing non-finite values (stat_summary).## Warning: Removed 7313 rows containing non-finite values (stat_summary).
## Warning: Removed 7313 rows containing non-finite values (stat_summary).## Warning: Removed 7313 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -155949.92 -2.44 0.05 -0.59 2.85 62730.15 5299
## 1% 99%
## -49.08547 51.33693## Warning: Removed 8097 rows containing non-finite values (stat_summary).## Warning: Removed 8097 rows containing non-finite values (stat_summary).
## Warning: Removed 8097 rows containing non-finite values (stat_summary).## Warning: Removed 8097 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -155949.92 -2.34 0.08 0.11 2.86 61821.25 3484
## 1% 99%
## -38.21115 42.13341## Warning: Removed 6282 rows containing non-finite values (stat_summary).## Warning: Removed 6282 rows containing non-finite values (stat_summary).
## Warning: Removed 6282 rows containing non-finite values (stat_summary).## Warning: Removed 6282 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -155949.92 -2.20 0.12 0.35 2.95 61821.25 2134
## 1% 99%
## -32.30106 38.89020## Warning: Removed 4898 rows containing non-finite values (stat_summary).## Warning: Removed 4898 rows containing non-finite values (stat_summary).
## Warning: Removed 4898 rows containing non-finite values (stat_summary).## Warning: Removed 4898 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -155949.92 -1.91 0.23 0.23 3.13 21406.72 989
## 1% 99%
## -25.72949 37.21585## Warning: Removed 3653 rows containing non-finite values (stat_summary).## Warning: Removed 3653 rows containing non-finite values (stat_summary).
## Warning: Removed 3653 rows containing non-finite values (stat_summary).## Warning: Removed 3653 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -300.3856 -0.2240 0.0534 0.1589 0.4502 255.1363 1277
## 1% 99%
## -3.542952 4.767548## Warning: Removed 4145 rows containing non-finite values (stat_summary).## Warning: Removed 4145 rows containing non-finite values (stat_summary).
## Warning: Removed 4145 rows containing non-finite values (stat_summary).## Warning: Removed 4145 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -148.2433 -0.1974 0.0741 0.2047 0.4917 255.1363 890
## 1% 99%
## -3.124281 4.696452## Warning: Removed 3730 rows containing non-finite values (stat_summary).## Warning: Removed 3730 rows containing non-finite values (stat_summary).
## Warning: Removed 3730 rows containing non-finite values (stat_summary).## Warning: Removed 3730 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -148.2433 -0.1563 0.1047 0.2607 0.5535 255.1363 573
## 1% 99%
## -2.731473 4.494093## Warning: Removed 3363 rows containing non-finite values (stat_summary).## Warning: Removed 3363 rows containing non-finite values (stat_summary).
## Warning: Removed 3363 rows containing non-finite values (stat_summary).## Warning: Removed 3363 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -148.24331 -0.09102 0.16278 0.35689 0.65655 108.68948 270
## 1% 99%
## -2.186726 4.207526## Warning: Removed 2944 rows containing non-finite values (stat_summary).## Warning: Removed 2944 rows containing non-finite values (stat_summary).
## Warning: Removed 2944 rows containing non-finite values (stat_summary).## Warning: Removed 2944 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -97.2506 -0.0332 0.0000 0.0066 0.0409 46.5893 1128
## 1% 99%
## -1.103962 1.161713## Warning: Removed 2958 rows containing non-finite values (stat_summary).## Warning: Removed 2958 rows containing non-finite values (stat_summary).
## Warning: Removed 2958 rows containing non-finite values (stat_summary).## Warning: Removed 2958 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -97.2506 -0.0291 0.0000 0.0050 0.0370 35.3041 961
## 1% 99%
## -0.8447286 0.8683545## Warning: Removed 2769 rows containing non-finite values (stat_summary).## Warning: Removed 2769 rows containing non-finite values (stat_summary).
## Warning: Removed 2769 rows containing non-finite values (stat_summary).## Warning: Removed 2769 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -97.2506 -0.0239 0.0000 0.0045 0.0331 35.3041 743
## 1% 99%
## -0.6312811 0.6441721## Warning: Removed 2517 rows containing non-finite values (stat_summary).## Warning: Removed 2517 rows containing non-finite values (stat_summary).
## Warning: Removed 2517 rows containing non-finite values (stat_summary).## Warning: Removed 2517 rows containing missing values (geom_point).
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -97.2506 -0.0176 0.0000 0.0062 0.0293 35.3041 510
## 1% 99%
## -0.4484312 0.4644258## Warning: Removed 2210 rows containing non-finite values (stat_summary).## Warning: Removed 2210 rows containing non-finite values (stat_summary).
## Warning: Removed 2210 rows containing non-finite values (stat_summary).## Warning: Removed 2210 rows containing missing values (geom_point).